2014 GSA Annual Meeting in Vancouver, British Columbia (19–22 October 2014)

Paper No. 288-15
Presentation Time: 11:45 AM


HOU, Xudong, Key Laboratory of Economic Stratigraphy and Palaeogeography, Nanjing Institute of Geology and Palaeontology, Chinese Academy of Sciences, No. 39 East Beijing Road, Nanjing, 210008, China, xdhou@nigpas.ac.cn

The fossil record preserves a wide range of events that might be used to build up timescales and correlate strata from place to place. The events include the originations and extinctions of species (or the FADs and LADs of species), the occurrences of distinctive faunal assemblages, magnetic field reversals, changes in ocean chemistry, and volcanic ash falls. A fundamental task of stratigraphy which involves a large amount of data collection and computation is to determine the regional or global sequence of all these events. It can hardly be achieved only by manual methods if without the help of modern database and computer technology. Quantitative stratigraphic software which is based on recent computing technology and numerical algorithm, can automatically sort, space and calibrate thousands of event data that are collected from hundreds of sections. Thus, this kind of software makes the reconstruction of high-resolution timescale possible. Presently, the major quantitative stratigraphic software includes SinoCor, CONOP, HA and so on.

CONOP (Constrained Optimization) is a piece of software which improves the algorithm of graphic correlation. It can correlate all the sections from multi-dimensional space by the simulated annealing algorithm which can find the global or local optimal solution of this kind of problem. But as the amount of data increases, the elapsed time of calculation will rise amazingly. For a normal-size data set which contains 500 sections and 10000 events, it will take as long as several months to compute the data set in CONOP, which is unacceptable for an ordinary research task. Thereby, a significant improvement in the computation power of the CONOP software is necessary.

With the rapid development of computer hardware, multicore computers are becoming more and more popular. This kind of computers possess the capacity of parallel computing, therefore, to implement the parallelization of CONOP will be evidently practical that can greatly improve the processing efficiency of CONOP. We adopted C# to rewrite the core code of CONOP9 which was originally written by Fortran and parallelized the most time-consuming functions. As a result, the parallelized CONOP is averagely five times faster than the original Fortran version of CONOP while carried out the same big data set.